Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
you going to deny you made these remarks?
also, go ahead and find a registered non-ecc memory for sale, try newegg or ebay, where are they?

You still can't read!

Please show me when I mentioned RDIMM non ECC memory's (this particular RAM) performance. Not RDIMM ECC vs UDIMM non ECC.

Also, please don't just crop a part of the post. I mentioned many times DRIMM can be faster when more than one DIMM installed on a single memory channel. You just can't read that.

Last but not least, I suggest you go back to your own thread, but not disturb this thread's discussion. You want to mess up your own thread, that's fine. But please respect the other members in this forum.
 
  • Like
Reactions: crjackson2134
you cannot install more than one DIMM of something that does not exist.

when will you admit you are wrong? they don't exist.
 
you cannot install more than one DIMM of something that does not exist.

when will you admit you are wrong? they don't exist.

I was talking about we can install more than one RDIMM ECC on a single memory channel. Can you understand this simple English sentence?

Again, please show me in which post I mentioned we can install more than one non ECC RDIMM? I never say that.
 
Last edited:
are you now saying that it's impossible to install more than one dimm? if it exists it should be possible. o_O
 
are you now saying that it's impossible to install more than one dimm? if it exists it should be possible. o_O

You are the only person said that, not me.

I only said "we can install more than one RDIMM ECC on a single memory channel" in post #29. Please feel free to prove me wrong.
 
it is either possible or not, make up your mind. you keep trying to weasel out of your statements. can you answer 3 simple questions without trying to dodge them...

#1 does it exist?
#2 where can we buy it?
#3 can we install more than one dimm of it? (if we can't even buy it, then no we can't)
 
AFAIK, UDIMM is faster only true when there is only one DIMM per channel. If installed two (or more) DIMM per channel, RDIMM can provide better memory bandwidth.
(notice he never mentions them being ECC or not being ECC)

you are right in that on the current market AFAIK there are no NON-ECC Registered DIMMs, however @h9826790 never specify mentions those initially...
 
(notice he never mentions them being ECC or not being ECC)

you are right in that on the current market AFAIK there are no NON-ECC Registered DIMMs, however @h9826790 never specify mentions those initially...

Considering the multi-channel RDIMM benefits detailed above,

Tech docs I’ve reviewed regarding x58 as well as the 4,1/5,1 single core setups. Memory slots will run triple channel when setup with 3 sticks of ram. With 4 slots filled, channel 3 will be combined/interleaved from slots 3/4.

So what happens if I run the ram as 8x16x16x16, giving channel 3 equal memory parts to combine? Massive slowdown in single and multi-core memory operations.

I’d consider upgrading to a x5690 to run 16x16x16x16 but $100 to access 8gb more of ram is a bit steep.
 
Considering the multi-channel RDIMM benefits detailed above,

Tech docs I’ve reviewed regarding x58 as well as the 4,1/5,1 single core setups. Memory slots will run triple channel when setup with 3 sticks of ram. With 4 slots filled, channel 3 will be combined/interleaved from slots 3/4.

So what happens if I run the ram as 8x16x16x16, giving channel 3 equal memory parts to combine? Massive slowdown in single and multi-core memory operations.

I’d consider upgrading to a x5690 to run 16x16x16x16 but $100 to access 8gb more of ram is a bit steep.
Massive perhaps in benchmarks but for the vast majority of applications the decrease wouldn't be perceptible.
 
  • Like
Reactions: ActionableMango
I’d consider upgrading to a x5690 to run 16x16x16x16 but $100 to access 8gb more of ram is a bit steep.

Just received and installed two 16GB 2Rx4 PC3L-10600R's (USD 100; eBay; used) to compliment my other six (different manu, but similar spec).

3x16GB/cpu

GB3: https://browser.geekbench.com/geekbench3/8644704
GB4: https://browser.geekbench.com/v4/cpu/8890001

4x16GB/cpu

GB3: https://browser.geekbench.com/geekbench3/8645952
GB4: https://browser.geekbench.com/v4/cpu/8922932

A discerned decrease in "bench" performance is noticeable when adding the additional 32GB. No noticeable performance decrease is obvious on the "desktop".

On a daily basis I normally utilize maybe 1/8 of the 96GB, and up to 1/4 when working with large documents, so this is entirely academic <smile>

I will probably purpose the 2x16GB in my T5500, as it can use the boost.

Regards, splifingate
 
121FC386-A945-4CE7-98C6-B949A95CE86A.jpeg
Interesting observation. Thanks for sharing.

With Dual Xeon CPU’s, there is a penalty for using 4 sticks per cpu. While using 4 sticks in a 990x, there is no penalty.

Considering the x5690 and the 990x are near identical twins, Could the slowdown come from:

  • ECC support on the x5690 vs none on the 990x
  • the northbridge struggling to address an additional 20 QPI lanes from the additional x5690
  • overhead of the two x5690 20 QPI lanes for Multi-cpu processing.
  • Variances in chipset behavior.
 
  • Like
Reactions: splifingate
View attachment 769463 Interesting observation. Thanks for sharing.

With Dual Xeon CPU’s, there is a penalty for using 4 sticks per cpu. While using 4 sticks in a 990x, there is no penalty.

Considering the x5690 and the 990x are near identical twins, Could the slowdown come from:

  • ECC support on the x5690 vs none on the 990x
  • the northbridge struggling to address an additional 20 QPI lanes from the additional x5690
  • overhead of the two x5690 20 QPI lanes for Multi-cpu processing.
  • Variances in chipset behavior.

I can't mismatch my processors on my trays, but someone with a 2010/2012 single/double tray could benchmark it. If I was betting, overhead on the QPI lanes.
 
I can't mismatch my processors on my trays, but someone with a 2010/2012 single/double tray could benchmark it. If I was betting, overhead on the QPI lanes.


keep in mind the Dual CPU Mac Pro 5,1 (and Xserve3,1) is NUMA not UMA, remember we are dealing with 2 memory controllers here

so yes if you have some data in memory thats connected via CPU 2 and CPU 1 requests it will you will end up bottle necked by the fact it has to go over the QPI link between the 2 CPUs to reach the 1st CPU

thats probably why some of the sub benchmarks in geekbench report much lower numbers on a Dual CPU 5,1 then they do on a Single CPU 5,1.
 
5,1 is NUMA?!? Awesome.

I opened Affinity Photo, and proceeded to conjure-up some memory-intensive ops (56Kpx/56Kpx + some tough back-history) which brought me up to ~24GB of ram usage; then I opened PS CC 2018 (latest), and commenced to tax it as far as I could, with the the same relative size image. . . IIRC, I could only broach ~60GB of App Memory with both, combined.

Bench #'s don't mean much, next to RW experience.

What would y'all advise to tax this system, to reveal the DP QPI inconsistencies?

Regards, splifingate
 
ECC has overhead. But the differences between all the results above in real world is negligible.

For fun comparison my PC has an overclocked 5GHZ 8700K and 4000MHZ quad channel memory. Geekbench score 6500/32000. That memory speed isn't supported natively by the CPU but it makes a noticeable difference when moving large data around in memory. This or better performance in coming to a Mac near you very soon.
 
ECC has overhead. But the differences between all the results above in real world is negligible.

For fun comparison my PC has an overclocked 5GHZ 8700K and 4000MHZ quad channel memory. Geekbench score 6500/32000. That memory speed isn't supported natively by the CPU but it makes a noticeable difference when moving large data around in memory. This or better performance in coming to a Mac near you very soon.

FYI an i7 8700K only has a Dual channel memory controller...
 
Following the link in the about this Mac memory panel. Apple's usual doc, along with Registered DIMM info.

window 7-7-185.39 PM.png

window 7-7-185.37 PM.png

window 7-7-185.37 PM-1.png
.
[doublepost=1531011819][/doublepost]As a scratch disk for any workflow, a RAMDisk's performance is touch to match on price / performance. $80 for a 32gb scratch drive with killer 4k performance is tough to beat.

Raid 0 ramdisk
2x2gb.png

vs a 970 pro
window7-1-186.57 PM.png
 
Last edited:
theres a big difference between Quad channel and just 2 DIMMs per channel on a dual channel memory controller keep in mind if you had a True Quad channel memory controller you would have double the memory bandwidth which is a big difference in some cases

Yes I know it's not a true quad channel. I would need an i9. I meant four sticks.
 
Sorry to bring this thread back -- I've just upgraded from a single CPU to dual CPU's on my 4,1->5,1.

If I'm hoping to rock dual x5690's, is the consensus that 6x16GB RDIMMS would be the best configuration? I'm trying to optimize a bit for memory latency and single threaded performance, but still want ECC.

Are UDIMMS a viable option, or are they limited by costs or the maximum memory I could configure?

Thanks.
 
I highly suggest 2rx4 Samsung ecc rdimms. pc3-12800 yko 16gb x 6 for 96gb. Using this chip I’ve scored the highest multi core results on Geekbench for an non overclocked 5677x dual 5677, dual 5680.

The ram down clocks to 1333mhz flawlessly

The best single core performance With dual x5690 you can expect is between 2900 and 3000 on a fresh install of macOS. Multi core should reach 26500 is for higher with the ram I specified.
 
  • Like
Reactions: eksu
I highly suggest 2rx4 Samsung ecc rdimms. pc3-12800 yko 16gb x 6 for 96gb. Using this chip I’ve scored the highest multi core results on Geekbench for an non overclocked 5677x dual 5677, dual 5680.

The ram down clocks to 1333mhz flawlessly

The best single core performance With dual x5690 you can expect is between 2900 and 3000 on a fresh install of macOS. Multi core should reach 26500 is for higher with the ram I specified.
Any idea what timings it’s running at 1333mhz?
 
  • Like
Reactions: handheldgames
Any idea what timings it’s running at 1333mhz?
They are rated as 9-9-9 at 1333.
[doublepost=1537142126][/doublepost]And should cost $44 ish each. Make sure you get 6 of the same chip. It's best to get it from one vendor. I've had great luck with sonypvm on ebay.
 
  • Like
Reactions: eksu
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.